arXiv subjects

Cuong Dang

Publications and source records attributed to Cuong Dang.

At least 19 recordsLinked to original sources

arXiv2026

Retrieved But Not Reliable: A Survey on Attacks, and Defenses in Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) enhances large language models by grounding outputs in external knowledge, improving factuality and reducing hallucinations. At the same time, the retrieval-augmented pipeline introduces new robustness and security risks, including corpus poisoning, backdoor attacks, privacy leakage, and fairness violations. Despite rapid progress in this area, existing surveys remain limited in their treatment of attacker objectives, threat models, and stage-specific defenses across the full RAG pipeline. This survey presents a unified and pipeline-aware overview of RAG robustness. We formalize threat models over the corpus, retriever, and generator, and organize attacks into three main objectives: accuracy, privacy, and fairness. We further review defenses from a pipeline-aware perspective, covering the retrieval, rerank, generation, and traceback stages. In addition, we summarize robustness benchmarks and explainability methods for more deeply evaluating and explaining RAG robustness.

cs.CR↗

arXiv2026

Capacity-Dependent Effects of Data Selection for Reasoning

In reasoning supervised fine-tuning, candidate responses for the same instruction can differ substantially in how well they match the student's current distribution. Recent likelihood-based response selection methods suggest that responses closer to the student distribution provide more effective supervision, motivating the hypothesis that high-likelihood responses may generally be preferable for fine-tuning. In this paper, we revisit this intuition and show that the value of likelihood-based data selection depends critically on model capacity and training duration. Through controlled experiments on mathematical reasoning, using students ranging from 1.5B to 8B parameters and supervision generated by stronger teacher models, we observe a clear \emph{capacity-dependent} ``{\color{SMALLCOLOR}\textbf{Fast-Fit}} / {\color{LARGECOLOR}\textbf{Slow-Gain}}'' pattern. High-likelihood data provides faster and more stable early improvements, especially for smaller models, but low-likelihood data becomes increasingly beneficial for larger models when training is allowed to continue longer. To explain this phenomenon, we analyze learning dynamics, showing that small models often fail to absorb low-likelihood supervision and instead fall into shallow or repetitive behaviors, while larger models are better able to move toward the teacher distribution under such data. We further provide a capacity-constrained theoretical view of distillation that clarifies how data difficulty, data span, and student capacity jointly govern transfer. Overall, our findings show that effective data selection for reasoning should be aware of model capacity and computing budget rather than based on a single universal preference for high-likelihood supervision.

cs.LG↗

arXiv2026

How Robust Are LLMs to Vietnamese Dialects?

Large Language Models (LLMs) are typically evaluated on standard written Vietnamese, yet everyday communication frequently involves regional dialects that preserve meaning but differ in surface form. Existing Vietnamese dialect work largely addresses this issue through dialect-to-standard normalization instead of measuring how the model fails under Vietnamese dialectal inputs. To address this gap, we present the first systematic evaluation of LLM robustness to Vietnamese dialect variation across multiple tasks, quantifying performance degradation and failure patterns. We introduce VialectBench (Vietnamese Dialects Benchmarking), a controlled benchmark for testing whether model decisions remain stable across six Vietnamese dialect groups. VialectBench contains 400 Standard Vietnamese source instances and 2,400 human-written dialectal rewrites spanning emotion recognition (ER), natural language inference (NLI), question answering (QA), and multiple-choice question answering (MCQA). Dataset evaluation with a fixed reference language model shows that the dialectal rewrites induce a measurable model-relative likelihood shift while remaining nearly equal in length to their Standard counterparts. Across ten instruction-tuned models, dialectal inputs reduce average performance by 2.82%, and no evaluated model is fully dialect-invariant. All four tasks are affected, with QA showing the largest average degradation. Robustness also varies substantially across dialect groups: PNT3 and PNT2 cause the largest average performance drops, at 6.17% and 4.73%, respectively, whereas PNB slightly improves average performance by 0.42%. The Central dialect group (PNT1-PNT4) also yields the highest average harmful-flip rate across all models, at 6.54%. These findings show that strong performance on Standard Vietnamese does not guarantee reliable behavior under meaning-preserving regional variation.

cs.CL↗

arXiv2026

The Confidence Trap: Calibration Attacks for Graph Neural Networks

While confidence calibration is essential for trustworthy decision-making in safety-critical applications, the robustness of calibrated GNNs to adversarial structural perturbations remains largely unexplored. However, studying calibration attacks on graphs presents unique technical challenges: (1) the discrete nature of graph structures complicates gradient-based optimization, (2) existing underconfidence objectives fail to drive predictions toward uniform distributions, and (3) GNNs are highly sensitive to edge perturbations, often causing unintended label changes that violate attack constraints. To address these challenges, we propose a \textbf{Unified Graph Calibration Attack (UGCA)} framework designed for \textbf{worst-case (white-box) analysis} of GNN calibration robustness. UGCA introduces a KL-divergence loss to encourage uniform predictive distributions, a reranking mechanism to reduce label flipping, a hybrid loss to recover labels when violations occur, and beam search to explore a broader adversarial search space. We further provide theoretical insights linking model generalization, dataset complexity, and calibration vulnerability, showing that models with higher accuracy or trained on datasets with more classes are more susceptible under this threat model. Extensive experiments demonstrate that UGCA substantially increases Expected Calibration Error while preserving classification accuracy. Our code is publicly available at https://github.com/CaptainCuong/Graph-Calibration-Attack.git.

cs.LG↗

arXiv2026

URAG: A Benchmark for Uncertainty Quantification in Retrieval-Augmented Large Language Models

Retrieval-Augmented Generation (RAG) has emerged as a widely adopted approach for enhancing LLMs in scenarios that demand extensive factual knowledge. However, current RAG evaluations concentrate primarily on correctness, which may not fully capture the impact of retrieval on LLM uncertainty and reliability. To bridge this gap, we introduce URAG, a comprehensive benchmark designed to assess the uncertainty of RAG systems across various fields like healthcare, programming, science, math, and general text. By reformulating open-ended generation tasks into multiple-choice question answering, URAG allows for principled uncertainty quantification via conformal prediction. We apply the evaluation pipeline to 8 standard RAG methods, measuring their performance through both accuracy and prediction-set sizes based on LAC and APS metrics. Our analysis shows that (1) accuracy gains often coincide with reduced uncertainty, but this relationship breaks under retrieval noise; (2) simple modular RAG methods tend to offer better accuracy-uncertainty trade-offs than more complex reasoning pipelines; and (3) no single RAG approach is universally reliable across domains. We further show that (4) retrieval depth, parametric knowledge dependence, and exposure to confidence cues can amplify confident errors and hallucinations. Ultimately, URAG establishes a systematic benchmark for analyzing and enhancing the trustworthiness of retrieval-augmented systems. Our code is available on GitHub.

cs.CL↗

arXiv2026

Topology-Enabled Switchable Unidirectional Radiative Band in a Bilayer Photonic Crystal

Controlling how an open photonic system exchanges energy with its environment-and in particular how it radiates into the far field-is a cornerstone of non-Hermitian wave physics and a key enabler for directional photonic functionalities. Here, we propose a new route to robust unidirectional emission based on the non-Hermitian hybridization of resonances localized in spatially separated layers of a hetero-bilayer photonic crystal. By tailoring the interlayer coupling, we engineer hybrid photonc bands that exhibit strong unidirectional radiation across a broad spectral and momentum range while maintaining theoretically high quality factors. This asymmetric emission is organized by a topological vortex in a pseudo-polarization field defined from the front/back intensity imbalance, which endows the directionality with robustness against perturbations. We further show that, by tuning the surrounding refractive index, this singularity can be displaced in parameter space, enabling reversible switching of the emission direction and a reconfigurable far-field response. This framework opens perspectives for topological photonic sensing and for directional and switchable light sources, including unidirectional lasing supported by high-quality-factor modes.

physics.optics↗

arXiv2025

Wavelength-scale noise-resistant on-chip spectrometer

Performant on-chip spectrometers are important for advancing sensing technologies, from environmental monitoring to biomedical diagnostics. As device footprints approach the scale of the operating wavelength, previously strategies, including those relying on multiple scattering in diffusive media, face fundamental accuracy constraints tied to limited optical path lengths. Here, we demonstrate a wavelength-scale, CMOS-compatible on-chip spectrometer that overcomes this challenge by exploiting inverse-designed quasinormal modes in a complex photonic resonator. These modes extend the effective optical path length beyond the physical device dimensions, producing highly de-correlated spectral responses. We show that this strategy is theoretically optimal for minimizing spectral reconstruction error in the presence of measurement noise. The fabricated spectrometer occupies a lateral footprint of only 3.5 times the free-space operating wavelength, with a spectral resolution of 10 nm across the 3.59-3.76 micrometer mid-infrared band, which is suitable for molecular sensing. The design of this miniaturized noise-resistant spectrometer is readily extensible to other portions of the electromagnetic spectrum, paving the way for lab-on-a-chip devices, chemical sensors, and other applications.

physics.optics↗

arXiv2025

Generalized Non-Hermitian Hamiltonian for Guided Resonances in Photonic Crystal Slabs

We develop a generalized non-Hermitian Hamiltonian formalism for guided resonances in photonic crystal slabs, derived directly from Maxwell's equations through a systematic guided-mode expansion. By expanding the electromagnetic fields over the complete mode basis of an unpatterned slab and systematically integrating out radiative Fabry--P\'erot channels, we obtain the analytical operator structure of the Hamiltonian, which treats guided-mode coupling and radiation losses on equal footing. The resulting Hamiltonian provides explicit expressions for both dispersive and radiative coupling terms in terms of modal overlap integrals and Fourier components of the permittivity modulation. For specific geometries, the Hamiltonian coefficients can be extracted from full-wave simulations enabling accurate modeling without phenomenological assumptions. As a case study, we investigate hexagonal lattices with both preserved and broken $C_6$ symmetry, demonstrating predictive agreement for complex band structures, near-field distributions, and far-field polarization patterns. In particular, the formalism reproduces symmetry-protected bound states in the continuum (BICs) at the $\Gamma$ point, accidental off-$\Gamma$ BICs near the $\Gamma$ point, and the emergence of chiral exceptional points (EPs). It also captures the tunable behavior of eigenmodes near the $K$ point, including Dirac-point shifts and the emergence of quasi-BICs or bandgap openings, depending on the nature of $C_6$ symmetry breaking. We further demonstrate in the Appendix that the same formalism extends naturally to other symmetry classes, including $C_2$ (1D grating) and $C_4$ (square lattice) photonic crystal slabs. This approach enables predictive and efficient modeling of complex photonic resonances, revealing their topological and symmetry-protected characteristics in non-Hermitian systems.

physics.optics↗

arXiv2025

Diffuse Optical Ptychography

Various imaging techniques have significantly enhanced our ability to visualize objects embedded within complex media such as biological tissues, fog, atmosphere, or various turbid media. Optical imaging, in particular, offers multiple advantages, including non-invasive capabilities, absence of ionizing radiation, and high contrast for many biological tissues. However, optical imaging through substantially thick scattering media remains challenging due to extensive photon diffusion, significantly restricting reconstruction quality and achievable resolution. To address these limitations, we introduce Diffuse Optical Ptychography (DOP), a novel imaging method inspired by ptychography technique, which exploits additional spatial information gained from multiple overlapping illumination patterns. The primary technical innovation of DOP lies in its effective use of overlapping yet minimally correlated illuminations, significantly enhancing reconstruction accuracy and image quality. Compared to existing optical imaging methods through thick diffusive media, DOP achieves superior resolution (down to 1 mm) and reliably reconstructs both binary and grayscale objects embedded within media thicker than 100 transport mean free paths. Importantly, DOP demonstrates robust reconstruction performance both with accurately calibrated diffusion properties and even without prior calibration. Furthermore, the experimental setup for DOP remains straightforward, utilizing only a conventional camera and scanning illumination spots. Our demonstrations underscore the broad potential impact of DOP in applications ranging from medical diagnostics to non-destructive testing, thus opening promising avenues for high-resolution imaging in highly scattering environments.

physics.optics↗

arXiv2025

Hybridization of Non-Hermitian Topological Interface Modes

We propose and experimentally demonstrate the hybridization of radiating topological interface states, analogous to Jackiw-Rebbi states but in gain media with radiation fields. This hybridization not only modifies energy levels under a strong coupling scheme but also significantly reshapes far-field radiation characteristics. The bonding mode exhibits sub-radiant, omnidirectional emission, while the antibonding mode becomes super-radiant and highly unidirectional. Crucially, this non-Hermitian hybridization is tunable, allowing simultaneous control of energy splitting, quality factor, and far-field radiation by varying the distance between the two topological interfaces. Our findings establish hybridized radiating topological interface states as a robust platform for engineering two-level systems with tailored far-field responses, offering new possibilities for applications in beam shaping, nonlinear optics, quantum technologies, and beyond.

physics.optics↗

arXiv2024

Scaling Up Purcell-Enhanced Self-Assembled Nanoplasmonic Perovskite Scintillators into the Bulk Regime

Scintillators convert high-energy radiation into detectable photons and play a crucial role in medical imaging and security applications. The enhancement of scintillator performance through nanophotonics and nanoplasmonics, specifically using the Purcell effect, has shown promise but has so far been limited to ultrathin scintillator films because of the localized nature of this effect. This study introduces a method to expand the application of nanoplasmonic scintillators to the bulk regime. By integrating 100-nm-sized plasmonic spheroid and cuboid nanoparticles with perovskite scintillator nanocrystals, we enable nanoplasmonic scintillators to function effectively within bulk-scale devices. We experimentally demonstrate power and decay rate enhancements of up to (3.20 $\pm$ 0.20) and (4.20 $\pm$ 0.31) folds for plasmonic spheroid and cuboid nanoparticles, respectively, in a 5-mm thick CsPbBr$_{3}$ nanocrystal-polymer scintillator at RT. Theoretical modeling also predicts similar enhancements of up to (2.26 $\pm$ 0.31) and (3.02 $\pm$ 0.69) folds for the same nanoparticle shapes and dimensions. Moreover, we demonstrate a (2.07 $\pm$ 0.39) fold increase in light yield under $^{241}$Am $\gamma$-excitation. These findings provide a viable pathway for utilizing nanoplasmonics to enhance bulk scintillator devices, advancing radiation detection technology.

physics.optics↗

arXiv2024

A Curious Case of Searching for the Correlation between Training Data and Adversarial Robustness of Transformer Textual Models

Existing works have shown that fine-tuned textual transformer models achieve state-of-the-art prediction performances but are also vulnerable to adversarial text perturbations. Traditional adversarial evaluation is often done \textit{only after} fine-tuning the models and ignoring the training data. In this paper, we want to prove that there is also a strong correlation between training data and model robustness. To this end, we extract 13 different features representing a wide range of input fine-tuning corpora properties and use them to predict the adversarial robustness of the fine-tuned models. Focusing mostly on encoder-only transformer models BERT and RoBERTa with additional results for BART, ELECTRA, and GPT2, we provide diverse evidence to support our argument. First, empirical analyses show that (a) extracted features can be used with a lightweight classifier such as Random Forest to predict the attack success rate effectively, and (b) features with the most influence on the model robustness have a clear correlation with the robustness. Second, our framework can be used as a fast and effective additional tool for robustness evaluation since it (a) saves 30x-193x runtime compared to the traditional technique, (b) is transferable across models, (c) can be used under adversarial training, and (d) robust to statistical randomness. Our code is publicly available at \url{https://github.com/CaptainCuong/RobustText_ACL2024}.

cs.LG↗

arXiv2021

Non-invasive super-resolution imaging through scattering media using fluctuating speckles

Extending super-resolution imaging techniques to objects hidden in strongly scattering media potentially revolutionize the technical analysis for much broader categories of samples, such as biological tissues. The main challenge is the media's inhomogeneous structures which scramble the light path and create noise-like speckle patterns, hindering the object's visualization even at a low-resolution level. Here, we propose a computational method relying on the object's spatial and temporal fluctuation to visualize nanoscale objects through scattering media non-invasively. The fluctuating object can be achieved by random speckle illumination, illuminating through dynamic scattering media, or flickering emitters. The optical memory effect allows us to derive the object at diffraction limit resolution and estimate the point spreading function (PSF). Multiple images of the fluctuating object are obtained by deconvolution, then super-resolution images are achieved by computing the high order cumulants. Non-linearity of high order cumulant significantly suppresses the noise and artifacts in the resulting images and enhances the resolution by a factor of $\sqrt{N}$, where $N$ is the cumulant order. Our non-invasive super-resolution speckle fluctuation imaging (NISFFI) presents a nanoscopy technique with very simple hardware to visualize samples behind scattering media.

physics.optics↗

arXiv2020

Hybrid plasmonic nano-emitters with controlled single quantum emitter positioning on the local excitation field

Hybrid plasmonic nanoemitters based on the combination of quantum dot emitters (QD) and plasmonic nanoantennas open up new perspectives in the control of light. However, precise positioning of any active medium at the nanoscale constitutes a challenge. Here, we report on the optimal overlap of antenna's near-field and active medium whose spatial distribution is controlled via a plasmon-triggered 2-photon polymerization of a photosensitive formulation containing QDs. Au nanoparticles of various geometries are considered. The response of these hybrid nano-emitters is shown to be highly sensitive to the light polarization. Different light emission states are evidenced by photoluminescence measurements. These states correspond to polarization-sensitive nanoscale overlap between the exciting local field and the active medium distribution. The decrease of the QD concentration within the monomer formulation allows trapping of a single quantum dot in the vicinity of the Au particle. The latter objects show polarization-dependent switching in the single-photon regime.

physics.optics↗

arXiv2020

The Emergence of Unconventional Plasmons Driven by Correlated Electron Interaction in B-site of 2D Hybrid Organic-Inorganic Perovskites

Hybrid organic-inorganic perovskites (HOIPs) have emerged to the forefront of optoelectronic materials advancement in the past few years. Due to the nature of organic compounds within the perovskite structure, its optoelectronic properties are affected by complex interaction and correlation effects between the organic and inorganic ions. Using spectroscopic ellipsometry, we observe two broad plasmonic excitation from the calculated loss function (LF) -Im[\varepislon^{-1} (ω)], peak A' and B' at 3.28 eV and 4.26 eV, respectively.The presence of these two asymmetric peaks in the spectroscopic ellipsometry (SE) spectra indicates the existence of unconventional plasmons at room temperature. This is inferred due to the absence of the zero-crossing in the real part of dielectric function \varepsilon_1 (ω). Through combined Near-Edge X-ray Absorption Fine Structure (NEXAFS) and Resonant Photoemission Spectroscopies (ResPES), we observe resonance enhancement peak close to 15 eV in the C K-edge region that unravels a charge transfer event due to the opening of an extra autoionization channel. Additionally, photoluminescence (PL) spectrum confirms the presence of broadband emission originating from the self-trapped emission excitons at 2.38 eV due to the soft 2D-HOIPs crystal structure. We believe that these phenomena directly impact the correlation strength in 2D-HOIPs. Our results have confirmed the existence of unconventional plasmons of 2D-HOIPs at room temperature. Such studies in the emission and plasmonic behavior of perovskites will pave the way for the efficient light emitting devices or lasers with minimal integrations of the materials.

cond-mat.mtrl-sci↗

arXiv2020

Non-invasive optical focusing inside strongly scattering media with linear fluorescence

Non-invasive optical focusing inside scattering media is still a big challenge because inhomogeneous media scatter both incoming photons for focusing and outgoing photons for observation. Various approaches, utilizing non-linear fluorescence or ultrasound, have been reported to address this difficulty. However, implementation of these methods is complicated and highly expensive, as ultrafast laser systems or photo-acoustic equipment must be employed. Here, we demonstrate a wavefront shaping technique to achieve non-invasive focusing (NiF) inside scattering media using only a linear fluorescent signal. Contrast and mean of incoherent speckles, produced by the linear fluorescence, are utilized as feedback signals to optimize the input wavefront. While increasing speckle contrast makes the focus tighter, and increasing the speckle mean enhances the intensity, fine-tuning the contribution of these two factors in our two-step optimization is essential. An optimal wavefront is found to achieve simultaneously both a micrometer focal spot size (down to 20 um diameter) and high intensity (more than a 100-fold enhancement) inside the scattering media. Our method promises a new route in life science towards focusing, imaging or manipulating deep into biological tissues with linear fluorescent agents.

physics.optics↗

arXiv2019

Manipulating Coherent Light Matter Interaction: Continuous Transition between Strong Coupling and Weak Coupling in MoS2 Monolayer Coupled with Plasmonic Nanocavities

Strong interactions between surface plasmons in ultra-compact nanocavities and excitons in two dimensional materials have attracted wide interests for its prospective realization of polariton devices at room temperature. Here, we propose a continuous transition from weak coupling to strong coupling between excitons in MoS2 monolayer and highly localized plasmons in ultra-compact nanoantenna. The nanoantenna is assembled by a silver nanocube positioned over a gold film and separated by a dielectric spacer layer. We observed a 1570-fold enhancement in the photoluminescence at weak coupling regime in hybrid nanocavities with thick spacer layers. The interaction between excitons and plasmons is then directly prompted to strong coupling regime by shrinking down the thickness of spacer layer. Room temperature formation of polaritons with Rabi splitting up to 190 meV was observed, which is the largest plasmon-exciton Rabi splitting reported in two dimensional materials. Numerical calculations quantified the relation between coupling strength, local density of states and spacer thickness, and revealed the transition between weak coupling and strong coupling in nanocavities. The findings in this work offer a guideline for feasible designs of plasmon-exciton interaction systems with gap plasmonic cavities.

cond-mat.mes-hall↗

arXiv2019

Noninvasive super-resolution imaging through scattering media

Super-resolution imaging with advanced optical systems has been revolutionizing technical analysis in various fields from biological to physical sciences. However, many objects are hidden by strongly scattering media such as rough wall corners or biological tissues that scramble light paths, create speckle patterns and hinder object's visualization, let alone super-resolution imaging. Here, we realize a method to do non-invasive super-resolution imaging through scattering media based on stochastic optical scattering localization imaging (SOSLI) technique. Simply by capturing multiple speckle patterns of photo-switchable emitters in our demonstration, the stochastic approach utilizes the speckle correlation properties of scattering media to retrieve an image with more than five-fold resolution enhancement compared to the diffraction limit, while posing no fundamental limit in achieving higher spatial resolution. More importantly, we demonstrate our SOSLI to do non-invasive super-resolution imaging through not only optical diffusers, i.e. static scattering media, but also biological tissues, i.e. dynamic scattering media with decorrelation of up to 80%. Our approach paves the way to non-invasively visualize various samples behind scattering media at unprecedented levels of detail.

physics.optics↗